Some of the organizations that benefit the most from using AI are in highly regulated industries like finance, government, and defense. Analyzing large volumes of sensitive records and preparing detailed reports are just some of tasks AI can help with.
But institutions in these industries need to control where AI runs, which information it can access, and what it’s allowed to do. Often that’s on the institutions’ own servers, not in the cloud.
In August, we launchedPortable Computer, which runs Perplexity Computer’s model and agents on hardware the user owns. It gives individuals a way to analyze private files and carry out multi-step work locally, with cloud access available when authorized.
Today we’re partnering with UK-based sovereign infrastructure company Valarian to bring Computer inside institutions’ own servers and systems. That includes air-gapped networks, which are isolated from the live web and other outside networks. Institutions can have Computer work on sensitive tasks while retaining control over the infrastructure, data access, and permissions governing that work.
Computer plans tasks, coordinates models and tools, and carries out research and multi-step workflows. Valarian’s ACRA platform runs it on infrastructure the institution controls and enforces the necessary access and network policies.
Computer across more environments
We’ve been expanding the ways people can work with Computer on their own hardware.Besides Portable Computer, Hybrid compute on Mac divides tasks between cloud models for planning, research, and reasoning and a local model for private files and sensitive actions. An on-device privacy gate controls what information may leave the Mac.
The Valarian partnership extends local deployment to institutional infrastructure. An estimated 45% of IT workloads still run in corporate data centers, according to Uptime Institute’s 2025 survey.
Organizations can deploy Computer across customer-controlled cloud environments and servers institutions physically own. This gives teams a way to bring research, analysis, and drafting into environments already governed by their organization’s security requirements.
Models and tools under institutional control
Computer brings together models, memory, tools, and isolated execution to carry out work using approved information. Institutions can run open-weight models such as Nemotron, GLM, Kimi, DeepSeek-R1, and Qwen on their own infrastructure, connecting them to the internal documents and systems needed for each task.
Different workloads can have different permissions. A market research agent could use public sources and approved external frontier models. Meanwhile, a legal agent could work with a local model and its assigned case files, with external connections blocked.
Valarian organizes these workloads into “enclaves,” which are controlled environments with defined access to data, tools, users, and credentials. Institutions set rules for which data and tools each workload can access, and the infrastructure enforces those limits.
Air-gapped deployments are even more controlled. They use the models, tools, and information available inside the isolated network. They operate without live connections to external model APIs, or the public web.
Sensitive work across industries
These deployment options support work that requires both detailed analysis and carefully scoped access. Examples include:
- Government: Prepare policy briefings from internal research and budget documents inside an isolated network, with access limited to approved agency records.
- Finance: Analyze confidential deal documents locally while a separate workload researches public markets through approved external connections.
- Legal: Build case chronologies and draft memos for attorney review, with access restricted to the assigned matter’s files and approved precedents.
- Healthcare: Summarize a patient’s medical history and recent test results for clinician review, with access restricted to authorized records inside the provider’s controlled environment.
Layered security controls
Deploying Computer with Valarian provides Valarian’s infrastructure controls with Computer’s isolated execution environments. Valarian’s platform governs access to data, tools, credentials, and network connections, keeping workloads with different permissions separate.
Computer uses sandboxes to contain code execution and file processing, with controlled access to the resources needed for a task. These protections help limit the reach of unexpected agent behavior.
Security teams get visibility into access decisions and workload activity with audit records. Valarian applies consistent controls across workloads, so institutions can expand Computer’s use without rebuilding the same safeguards for each application.
Deployments available now
Computer with Valarian is built for regulated institutions and other organizations that need advanced AI capabilities while keeping sensitive data inside their own systems.
Deployments are available starting today. Email sovereignai@perplexity.ai to discuss a deployment.
